BULLETIN (New Series) OF THEAMERICAN MATHEMATICAL SOCIETYVolume 29, Number 1, July 1993
"THEORETICAL MATHEMATICS": TOWARD A CULTURAL
SYNTHESIS OF MATHEMATICS AND THEORETICAL PHYSICS
ARTHUR JAFFE AND FRANK QUINN
Abstract. Is speculative mathematics dangerous? Recent interactions between
physics and mathematics pose the question with some force: traditional mathe-
matical norms discourage speculation, but it is the fabric of theoretical physics.
In practice there can be benefits, but there can also be unpleasant and destructive
consequences. Serious caution is required, and the issue should be considered
before, rather than after, obvious damage occurs. With the hazards carefully in
mind, we propose a framework that should allow a healthy and positive role for
speculation.
Modern mathematics is nearly characterized by the use of rigorous proofs.
This practice, the result of literally thousands of years of refinement, has brought
to mathematics a clarity and reliability unmatched by any other science. But it
also makes mathematics slow and difficult; it is arguably the most disciplined
of human intellectual activities.
Groups and individuals within the mathematics community have from time
to time tried being less compulsive about details of arguments. The results have
been mixed, and they have occasionally been disastrous. Yet today in certain
areas there is again a trend toward basing mathematics on intuitive reasoning
without proof. To some extent this is the old pattern of history being repeated
by those unfamiliar with it. But it also may be the beginning of fundamental
changes in the way mathematics is organized. In either case, it is vital at this
time to reexamine the role of proofs in mathematical understanding and to
develop a constructive context for these trends.
We begin with a discussion of physics, partially because some of the current
movement results from an interaction with theoretical physics and partly be-
cause it provides a useful model for potential sociological realignments. We
then turn to the history of mathematics for examples illustrating the benefits
and hazards of nonrigorous work. Finally, we suggest a framework which should
allow different approaches to coexist.
Theory and rigor
Typically, information about mathematical structures is achieved in two
stages. First, intuitive insights are developed, conjectures are made, and specu-
lative outlines of justifications are suggested. Then the conjectures and specu-
lations are corrected; they are made reliable by proving them. We use the term
theoretical mathematics for the speculative and intuitive work; we refer to the
proof-oriented phase as rigorous mathematics. We do not wish to get involved
in a discussion of our choice of terminology; this is not the central point of our
Received by the editors November 6, 1992.
1991 Mathematics Subject Classification. Primary 01A80.Both authors were partially supported by the National Science Foundation.
©1993 American Mathematical Society0273-0979/93 $1.00+ $.25 per page
1
2 ARTHUR JAFFE AND FRANK QUINN
article. However, as a point of departure, we briefly clarify our definitions of
theory and rigor.The initial stages of mathematical discovery—namely, the intuitive and con-
jectural work, like theoretical work in the sciences—involves speculations on
the nature of reality beyond established knowledge. Thus we borrow our name
"theoretical" from this use in physics. There is an older use of the word "theo-
retical" in mathematics, namely, to identify "pure" rather than applied mathe-
matics; this is a usage from the past which is no longer common and which we
do not adopt.Theoretical work requires correction, refinement, and validation through ex-
periment or proof. Thus we claim that the role of rigorous proof in mathematics
is functionally analogous to the role of experiment in the natural sciences. This
thesis may be unfamiliar but after reflection should be clear at least to math-
ematicians. Proofs serve two main purposes. First, proofs provide a way to
ensure the reliability of mathematical claims, just as laboratory verification pro-
vides a check in other sciences. Second, the act of finding a proof often yields,
as a byproduct, new insights and unexpected new data, just as does work in the
laboratory.Mathematicians may have even better experimental access to mathematical
reality than the laboratory sciences have to physical reality. This is the point of
modeling: a physical phenomenon is approximated by a mathematical model;
then the model is studied precisely because it is more accessible. This accessibil-
ity also has had consequences for mathematics on a social level. Mathematics
is much more finely subdivided into subdisciplines than physics, because the
methods have permitted a deeper penetration into the subject matter.
Although the use of proof in mathematics is functionally parallel to experi-
ment, we are not suggesting that proofs should be called "experimental" math-
ematics. There is already a well-established and appropriate use of that term,
namely, to refer to numerical calculations and computer simulations as tests of
mathematical concepts. In fact, results of computer experiments are frequently
presented in a way we would call theoretical: general conclusions are proposed
on the basis of test cases and comparisons. The conclusions are not completelyreliable, and an effort to provide real proofs would inevitably turn up exceptions
and limitations.
Division of labor
In physics we have come to accept a division of labor between theorists and
experimentalists. But in fact only recently has the division become so clear.Until the beginning of the twentieth century there was basically one community
of physicists. It was the ideal, and by and large the practice, that the same peopleboth speculated about theory and verified their speculations with laboratory
experience.Certainly in Europe it had become clear by 1900 that a bifurcation had oc-
curred: there were sufficiently many physicists who concentrated solely on the
theoretical side of their work that one could identify two distinct communi-
ties [H]. This trend proceeded somewhat more slowly in the United States.
E. C. Kemble, who worked at Harvard in the area of quantum theory, is gen-
erally regarded as the first American to obtain a doctorate in purely theoreticalphysics (though his 1917 thesis contained an experimental appendix).
THEORETICAL MATHEMATICS 3
Distinctness of the theoretical and experimental physics communities should
not be confused with their independence. Theory is vital for experimentalists to
identify crucial tests and to interpret the data. Experiment is vital for theorists
to correct and to guide their speculations. Theoretical and experimental groups
are unstable and ineffective unless they occur in closely interacting pairs.
We contrast this division of labor in physics to the current situation in math-
ematics. It is still the ideal, and by and large the practice, that the same peo-
ple both speculate about mathematical structures and verify their speculations
through rigorous proofs. In other words, the mathematical community has not
undergone a bifurcation into theoretical and rigorous branches. Even on the
individual level we seldom recognize theoretical mathematics as an appropriate
principal activity.There have been attempts to divide mathematical efforts in this way, but these
attempts were for the most part unsuccessful. Why? Is there something about
mathematics itself which invalidates the analogy with physics and prevents such
a bifurcation? Is "theoretical mathematics" in the end an oxymoron? Or were
there flaws in past attempts which doomed them but which might be avoided
today?
New relations with physics
A new connection with physics is providing a good deal of the driving force to-
ward speculation in mathematics. Recently there has been a flurry of mathemati-
cal-type activity in physics, under headings like "string theory", "conformai field
theory", "topological quantum field theory", and "quantum gravity". In large
part this has been initiated by individuals trained as theoretical high-energy
physicists. The most celebrated and influential of these (though not the most
problematic) is Edward Witten.From a physical point of view much of this work has not yet matured to
the stage where observable predictions about nature have been made. Further,
the work often concerns "toy models" designed to display only analogs of real
phenomena. And some of the parts which might apply to the real world concern
experimentally inaccessible events: particles of incredible energy, movement on
the scale of the universe, or creation of new universes.
One result of the lack of predictions is that these physicists are cut off from
their presumptive experimental community; they have no source of relevant
physical facts to constrain and inspire their theorizing. Since progress comes
from the interaction between theory and experiment, a theoretical group cannot
exist long in isolation. Indeed much of the mainstream physics community
regards these developments with suspicion, because of their isolation from the
so-called "real world".But these physicists are not in fact isolated. They have found a new "exper-
imental community": mathematicians. It is now mathematicians who provide
them with reliable new information about the structures they study. Often it is
to mathematicians that they address their speculations to stimulate new "exper-
imental" work. And the great successes are new insights into mathematics, not
into physics. What emerges is not a new particle but a description of represen-
tations of the "monster" sporadic group using vertex operators in Kac-Moody
algebras. What is produced is not a new physical field theory but a new view
of polynomial invariants of knots and links in 3-manifolds using Feynman path
integrals or representations of quantum groups.
4 ARTHUR JAFFE AND FRANK QUINN
These physicists are still working in the speculative and intuitive mode of the-
oretical physics. Many have neither training for nor interest in rigor. They aredoing theoretical mathematics. Very senior mathematicians have praised this
work and have suggested it should be emulated. As a result, some mathematical
followers are moving toward a more speculative mode.
One could conclude from this description that parts of mathematics have
already been propelled through the bifurcation. These areas have suddenly
acquired a fully functioning theoretical community, the theoretical physicists,
and traditional mathematicians have become the partner community of rigorous
verifiers.However, this has happened without the evolution of the community norms
and standards for behavior which are required to make the new structure stable.
Without rapid development and adoption of such "family values" the new rela-
tionship between mathematics and physics may well collapse. Physicists will go
back to their traditional partners; rigorous mathematicians will be left with a
mess to clean up; and mathematicians lured into a more theoretical mode by the
physicists' example will be ignored as a result of the backlash. To understand
what is involved, we start with a look at the past.
OLD RELATIONS WITH PHYSICS
The rich interplay between mathematics and physics predates even their
recognition as separate subjects. The mathematical work that in some sense
straddles the boundaries between the two is commonly referred to as mathe-matical physics, though a precise definition is probably impossible. In particular
over the past ninety years or so, a school of mathematical physicists emerged
around such persons as D. Hubert, F. Klein, H. Poincaré, M. Born, and laterH. Weyl, J. von Neumann, E. P. Wigner, M. Kac, A. S. Wightman, R. Jost,and R. Haag—persons with training both in physics and in mathematics. These
people often worked on questions motivated by physics, but they retained thetraditions and the values of mathematics.
Results emerging from this school have at various times been relevant both for
mathematics and for physics. Their development has proceeded at a deliberate
pace with the accumulation of conclusions of long-term interest. A few (out of
many) recent examples include work on the existence of quantum field theory
and on its compatibility with relativity; the work of Lieb and of Baxter on
lattice models and related transfer matrices; the work of Schoen and Yau on the
positive energy theorem in relativity and its relation to the geometry of minimal
surfaces; the work of Ruelle and others on dynamical systems and turbulence;
the operator algebra approach to local quantum theory; and Connes's early
interest in physics, from which emerged his mathematical work on factors and
later the foundation of noncommutative geometry.
The main point here is what has not happened. Work from this school is
characterized by standards of scholarship and by knowledge of the literature
consistent with the best traditions of mathematics. There is no ambiguity about
definitions, the formulation of claims, or proofs of theorems. It is traditional
rigorous mathematics. Even in this close proximity to physics the traditionalvalues of mathematics have been retained. Mathematical physicists do have
access to a vast, rich body of speculation by theoretical physicists. But these
THEORETICAL MATHEMATICS 5
speculations have traditionally been addressed to physicists, not mathemati-
cians.Theoretical physics and mathematical physics have rather different cultures,
and there is often a tension between them. Theoretical work in physics does
not need to contain verification or proof, as contact with reality can be left
to experiment. Thus the sociology of physics tends to denigrate proof as an
unnecessary part of the theoretical process. Richard Feynman used to delight
in teasing mathematicians about their reluctance to use methods that "worked"
but that could not be rigorously justified [F, G2]. He felt it was quite satisfactory
to test mathematical statements by verifying a few well-chosen cases.
On the mathematical side, E. J. McShane once likened the reasoning in a
"physical argument" to that of "the woman who could trace her ancestry to
William the Conqueror with only two gaps", and this was typical of the math-
ematical attitude. There was an understanding that the development must pro-
ceed at an internally appropriate pace and not be stampeded by the extent of the
physicists' vision. McShane and most of the mathematical physics community
rejected the free-wheeling, theoretical approach as inappropriate.
A relevant observation is that most theoretical physicists are quite respectful
of their experimental counterparts. Relations between physics and mathemat-
ics would be considerably easier if physicists would recognize mathematicians
as "intellectual experimentalists" rather than think of them disdainfully as use-
lessly compulsive theorists. The typical attitude of physicists toward mathemat-ics is illustrated by a passage from a book of P. W. Anderson, "We are talking
here about theoretical physics, and therefore of course mathematical rigor is
irrelevant and impossible."1 In fact it is exactly as relevant and possible as ex-
perimental data, and like data should be used whenever available. Nevertheless,
students in physics are generally indoctrinated with antimathematical notions;
and if they become involved in mathematical questions, they tend not only to
be theoretical but often to deny that their work is incomplete.
Not all of mathematical physics has been as clearly mathematical as that
described above. For example, the work of the German mathematical physicistK. Symanzik was mostly theoretical. However, he was very careful not to make
unwarranted claims of mathematical rigor. In fact, he made a serious attemptin 1968 to establish an important part of his theoretical program on a rigorouslevel in collaboration with the mathematician S. R. S. Varadhan. Some years
afterward, this was achieved by E. Nelson, Osterwalder, Schrader, and others
and had far-reaching consequences relating quantum theory, probability, and
statistical mechanics.
We also mention two people who have been addressing largely mathematical
questions from an almost entirely theoretical point of view for some time—
namely, B. Mandelbrot and M. Feigenbaum. See [G] for a popular account and
[Kr] for an expression of the mathematical discomfort with this activity.
For the most part the mainstream of mathematical physics has rejected purely
theoretical work as a valid mathematical style. We observe, however, that the
mathematicians involved in the "new relations" with physics are different from
'This is attributed by Anderson to Landau [A, p. 132]. Anderson continues, "This is not quite
so, but it is very close to it." However, he revealingly remembered the passage incorrectly; it reads
[LL], "No attempt has been made at mathematical rigor in the treatment, since this is anyhow
illusory in theoretical physics,... ."
6 ARTHUR JAFFE AND FRANK QUINN
the traditional mathematical physicists. Geometers, topologists, and persons in
representation theory have begun to talk with physicists. These mathematicians
are unused to dealing with the difference in cultures and for the most part do
not recognize parallels in their own collective experience that would sensitize
them to the hazards of theoretical work. This suggests a question: As they gain
more experience, will these mathematicians also reject pure theorizing?
Success stories
We turn to mathematical encounters with theoretical work and begin with
the positive side. The posing of conjectures is the most obvious mathematical
activity that does not involve proof. Conjectures range from brilliant to boring,
from impossible to obvious. They are filtered by the interest they inspire rather
than by editors and referees. The better ones have inspired the development of
whole fields. Some of the most famous examples are the Riemann Hypothesis,
Fermat's "last theorem", and the Poincaré conjecture. The Hilbert problem list,
of amazing breadth and depth, has been very influential in the development of
mathematics in this century. Other examples are the Adams conjecture in topol-
ogy, the several Questions of Serre, the Novikov conjecture, and the Wightman
axioms for quantum field theory.
Some conjectures are accompanied by technical details or a proposal for a
proof. For example, the "Weil conjectures" outlined an approach to a p-adic
analog of the Riemann hypothesis. The implementation of this program by
Grothendieck and Deligne was celebrated as a major achievement for modern
algebraic geometry. Similarly celebrated was Falting's proof of the Mordell con-
jecture, part of a program to approach the Fermât "theorem". The "Langlands
program" for understanding automorphic forms has been a major stimulus to
that field, and "Mori's program" for the investigation of algebraic three-folds
has invigorated that area. The classification of finite simple groups was achieved
by implementing a program developed by Gorenstein.
These examples share the characteristic that they were explicitly speculative
when formulated (or at least quickly recognized as speculative, as for example
with Fermât). They represented a goal to work toward, and primary credit for
the achievement was clearly to be assigned to the person who found a proof.
Another type of mathematical work is intermediate between traditional and
theoretical. It proceeds in the way, "If A is true, then X, Y, and Z follow", or
"If A, then it is reasonable to conjecture R, S, and T." In this case "A" may
be a major outstanding mathematical conjecture, the Riemann Hypothesis for
instance. A striking recent example of work in this style occurs in the theory
of motifs, and recently an entire Séminaire Bourbaki was devoted to the evolu-
tion of work in number theory and algebraic geometry based on conjectures ofDeligne and of Beilinson [Fo]. It is interesting that the Bourbaki, once the bas-
tion of the most conservative traditional mathematics, now entertains pyramidsof conjectures.
The importance of large-scale, goal-formulating work (necessarily theoretical)
is growing. We are in an age of big science, and mathematics is not an exception.
The classification of finite simple groups, for instance, is estimated to occupy
15,000 journal pages! Other sciences have responded to this trend by forminglarge formal collaborative groups.
The National Science Foundation and government agencies in other countries
THEORETICAL MATHEMATICS 7
have tried to nudge mathematics in the direction of collaborative and interdisci-plinary work. But one must recognize that large projects in mathematics are not
centered around a grant, a technique, or a machine. They are undertaken by in-
formal communities nucleated by a visionary theoretical program. In the finite
simple group effort, the program was developed and coordinated by D. Goren-
stein, who parcelled out pieces of the puzzle to an informal "team" working on
the problem. Future growth of such large-scale mathematical activity can only
occur with the evolution of more such visionary programs.
Cautionary tales
Most of the experiences with theoretical mathematics have been less positive
than those described above. This has been particularly true when incorrect or
speculative material is presented as known and reliable, and credit is claimed by
the perpetrator. Sometimes this is an "honest mistake", sometimes the result of
nonstandard conceptions of what constitutes proof. Straightforward mistakes
are less harmful. For example, the fundamental "Dehn lemma" on two-disks in
three-manifolds was presented in 1910. An error was found, and by the time
it was proved (by Papakyriakopolos in 1957) it was recognized as an importantconjecture.
Weak standards of proof cause more difficulty. In the eighteenth century,
casual reasoning led to a plague of problems in analysis concerning issues like
convergence of series and uniform convergence of functions. Rigor was intro-
duced as the antidote. It was adopted over the objections of some theorists intime to avoid major damage.
More recently in this century the "Italian school" of algebraic geometry did
not avoid major damage: it collapsed after a generation of brilliant speculation.
See [EH, K] for discussions of the difficulties and the long recovery. In 1946 the
subject was still regarded with such suspicion that Weil felt he had to defendhis interest in it; see the introduction to [W].
Algebraic and differential topology have had several episodes of excessively
theoretical work. In his history [D], Dieudonné dates the beginning of the field
to Poincaré's Analysis Situs in 1895. This "fascinating and exasperating paper"
was extremely intuitive. In spite of its obvious importance it took fifteen or
twenty years for real development to begin. Dieudonné expresses surprise at
this slow start [D, p. 36], but it seems an almost inevitable corollary of how it
began: Poincaré claimed too much, proved too little, and his "reckless" methods
could not be imitated. The result was a dead area which had to be sorted outbefore it could take off.
Dieudonné suggests that casual reasoning is a childhood disease of mathe-
matical areas and says, "... after 1910... uniform standards of what constitutesa correct proof became universally accepted in topology... this standard has
remained unchanged ever since." But in fact there have been many further
episodes. René Thorn's early work on differentiable manifolds, for which he
received the Fields medal, was brilliant and generally solid. Later work on sin-
gularities was not so firm. His claim of C°° density of topologically stable maps
was supported by a detailed but incomplete outline, which was later repaired
by John Mather. Thorn went on to propose the use of "catastrophe theory",founded on singularities, to explain forms of physical phenomena. The appli-
cation was mathematically theoretical, and its popularization, particularly by
8 ARTHUR JAFFE AND FRANK QUINN
E. C. Zeeman, turned out to be physically controversial.
The early work of Dennis Sullivan provides another example. After a solid
beginning, in the 1970s he launched into a brilliant and highly acclaimed but
"theoretical" exploration of the topology of manifolds. Details were weak, and
serious efforts to fill them in got bogged down. Sullivan himself changed fields
and returned to a more rigorous approach. This field still seems to have more
than its share of fuzzy proofs.
William Thurston's "geometrization theorem" concerning structures on
Haken three-manifolds is another often-cited example. A grand insight deliv-
ered with beautiful but insufficient hints, the proof was never fully published.
For many investigators this unredeemed claim became a roadblock rather than
an inspiration.
In these examples, as with Poincaré, the insights proposed seem to be on
target. There are certainly cases in which the theoretical insights were also
flawed. The point is that even in the best cases there were unpleasant side effects
that might have been avoided. We also see that Witten, in giving a heuristic
description of an extension of the Jones polynomial [Wi], was continuing in a
long and problematic tradition even within topology.
Some areas in the Russian school of mathematics have extensive traditions
of theoretical work, usually conducted through premature research announce-
ments. From the numerous possible examples we mention only two. The first is
concerned with the perturbation theory of integrable Hamiltonian systems with
phase space foliated by invariant tori. In 1954 Kolmogorov announced thattori with nonresonant frequencies survive a perturbation and gave an outline
of an argument. In retrospect it may be seen that this outline does touch on
the major ideas necessary, but it was generally considered insufficient to allow
reconstruction of a proof. Complete proofs were achieved by Arnold in 1959
for the analytic case and by Moser in 1962 for the smooth case.
The second example is one in which one of us became personally involved
while working on trying to establish a widely conjectured result that phase tran-
sitions occur in (relativistic) quantum field theory. In 1973 the respected math-
ematicians Dobrushin and Minios published an announcement of that result.
Two years later when no indication had come from the Russians of a proof,
Glimm, Jaffe, and Spencer resumed their work on the problem and eventually
gave two different proofs. A couple of years after that Dobrushin and Minios
published a retraction of their original announcement.
The problems
There are patterns in the problems encountered in these examples. We list
some and then discuss them in more detail.
(1) Theoretical work, if taken too far, goes astray because it lacks the feedbackand corrections provided by rigorous proof.
(2) Further work is discouraged and confused by uncertainty about whichparts are reliable.
(3) A dead area is often created when full credit is claimed by vigorous
theorizers: there is little incentive for cleaning up the debris that blocks furtherprogress.
(4) Students and young researchers are misled.
The first problem often overtakes would-be mathematical theorizers, particu-
THEORETICAL mathematics 9
larly when they are unwilling to acknowledge that their work is uncertain and
incomplete. Even in theoretical physics where there is an awareness of this pos-
sibility, knowing when to stop is a subtle and difficult skill. Errant theorizing
damages the credibility of the theorist and may also damage the field through
the mechanism identified in the second problem.
The second problem has to do with uncertainty of the literature. In compar-
ison with other sciences, the primary mathematical literature is extraordinarilyreliable. Papers in refereed core mathematics journals are nearly always sound,
and this permits steady and efficient advance. Only a small pollution of serious
errors would force mathematicians to invest a great deal more time and energy
in checking published material than they do now. The advantages of a reli-able literature are so profound that we suspect this is the primary force driving
mathematics toward rigor.When reliability of a literature is uncertain, the issue must be addressed. Of-
ten "rules of thumb" are used. For example, mathematicians presume that pa-
pers in physics journals are theoretical. This extends to a suspicion of mathema-
tical-physics journals, where the papers are generally reliable (though with dan-
gerous exceptions). Another widely applied criterion is that anything using func-
tional integrals must be speculative. One of us has remarked on the difficulties
this causes mathematicians trying to use solid instances of the technique [J].
These kinds of rules are unsatisfactory, as is the caveat emptor approach of let-
ting each paper be judged for itself. Proponents of this latter view cite Witten's
papers as successful examples. But a few instances can be handled; it is large
numbers that are a disaster. Also, it is a common rule of thumb now to regard
any paper by Witten as theoretical. This short-changes Witten's work but illus-
trates the "better-safe-than-sorry" approach mainstream mathematics tends to
take when questions arise.This unreliability is certainly a problem in theoretical physics, where the
primary literature often becomes so irrelevant that it is abandoned wholesale.
I. M. Singer has compared the physics literature to a blackboard that must be
periodically erased. Physicists traditionally obtain much less benefit from the
historical background of a problem, and they are less apt to search the literature.The citation half-life of physics papers is much shorter than in mathematics.
The "dead area" problem concerns credit and rewards. Mathematical re-
searchers traditionally do not give credit twice for the same results. But this
means that when a theorist claims credit, it is difficult for rigorous workers tojustify the investment of labor required to make it reliable. There is a big dif-
ference between "filling in the details of a theorem by X " and "verifying a
conjecture of X ". Rigorous mathematicians tend to flee the shadow of a big
claim. The pattern is that the missing work is filled in, often much later, using
techniques and corollaries of work on separate topics for which uncontested
credit is available.Finally, on the last point most successful theorizers (at least in mathemat-
ics) have a solid background in disciplined work, which is the source of their
intuition and taste. Most students who try to dive directly into the heady world
of theory without such a background are unsuccessful. Failure to distinguish
between the two types of activity can lead students to try to emulate the more
glamorous and less disciplined aspects and to end up unable to do more than
manipulate jargon.
10 ARTHUR JAFFE AND FRANK QUINN
Mathematicians tend to focus on intellectual content and neglect the impor-
tance of social issues and the community. But we are a community and often
form opinions even on technical issues by social interactions rather than directly
from the literature. Socially accepted conventions are vital in our understanding
of what we read. Behavior is important, and the community of mathematicians
is vulnerable to damage from inappropriate behavior.
Prescriptions
The mathematical community has evolved strict standards of proof and
norms that discourage speculation. These are protective mechanisms that guard
against the more destructive consequences of speculation; they embody the col-
lective mathematical experience that the disadvantages outweigh the advantages.
On the other hand, we have seen that speculation, if properly undertaken, can be
profoundly beneficial. Perhaps a more conscious and controlled approach that
would allow us to reap the benefits but avoid the dangers is possible. The need
to find a constructive response to the new influences from theoretical physics
presents us with both an important test case and an opportunity.
Mathematicians should be more receptive to theoretical material but with
safeguards and a strict honesty. The safeguards we propose are not new; they
are essentially the traditional practices associated with conjectures. However,
a better appreciation of their function and significance is necessary, and they
should be applied more widely and more uniformly. Collectively, our proposals
could be regarded as measures to ensure "truth in advertising".
Theoretical work should be explicitly acknowledged as theoreti-
cal and incomplete; in particular, a major share of credit for the
final result must be reserved for the rigorous work that validates
it.
This can make the difference between a dead area and a living industry. The-
oreticians should recognize that in the long run the success of their work is
dependent on the work of a companion rigorous community; they should honor
and nurture it when possible. Certainly in physics the community assigns ba-
sic credit for discovery to successful experimental investigations; they are not
regarded merely as verifying small details in the web of theoretical insight. Onthe individual level mathematical authors should make a choice: either they
provide complete proofs, or they should agree that their work is incomplete
and that essential credit will be shared. Referees and editors should enforce
this distinction, and it should be included in the education of students.
The other suggestions are concerned with the integrity of the mathematical
literature. It has always been acceptable to state a conjecture in a paper, and
occasionally papers have been published that are entirely theoretical. The key
issue is to identify the theoretical material clearly.
Within a paper, standard nomenclature should prevail: in the-
oretical material, a word like "conjecture" should replace "the-
orem"; a word like "predict" should replace "show" or "con-
struct"; and expressions such as "motivation" or "supporting
argument" should replace "proof. Ideally the title and abstract
should contain a word like "theoretical", "speculative", or "con-
jectural".
theoretical mathematics 11
The objective is to have flags indicating the nature of the work to readers. A
flag in the title of a theoretical paper would appear in a citation, which would
help limit second-hand problems. Theoretical work should be cited as a source
of inspiration or to justify significance or in supporting arguments in other
theoretical developments. Citing a theoretical paper for a structural ingredient
of a supposedly rigorous proof must be handled with care, and a flag in the title
would indicate when such care is needed.
Research announcements pose particular problems. Some announcements
are simply summaries of work that the author has completed and written down.
In such announcements, the language of theorem and proof is appropriate. Oth-
ers describe outcomes of arguments which have not been worked out in detail,
and sometimes they contain leaps of faith which require years of effort to bridge.
In these cases the traditional language misrepresents the work and is not appro-
priate; such announcements really should be identified as theoretical. The anal-ysis above implies that because of this, publication of announcements may be
unwholesome or even damaging to the fabric of mathematics. Announcements
do have valid functions, for instance, establishing a claim to priority and alert-ing others to new results and useful techniques. We therefore seek guidelines, as
with theoretical work in general, which will permit beneficial uses but limit the
potential damage. The key issue seems to be incorporation into the literature.One solution is:
Research announcements should not be published, except as
summaries of full versions that have been accepted for publica-
tion. Citations of unpublished work should clearly distinguish
between announcements and complete preprints.
In this age of copying machines and electronic bulletin boards it is possible to
distribute information widely without formally publishing it. "Cross-cultural"
preliminary results of wide interest could be described informally in "news col-
umn" format in publications like the Mathematical Intelligencer or the Notices
of the American Mathematical Society. The lack of formal publication, there-
fore, should be at most a minor disadvantage. We observe that this discussion
also underscores the importance of maintaining the distinction between formal
(refereed) publication and a posting on a bulletin board. Maintaining this dis-
tinction may be one of the greatest challenges facing the development of seriouselectronic journals.
If these safeguards are carefully followed, then it would be reasonable for any
mathematics journal to consider theoretical articles for publication. Stimulating
articles with incomplete proofs might be offered publication (after appropriate
word changes) as theoretical papers. Theoretical mathematics journals might
be appropriate. But care is required. Without honesty and caution by authors,editors, and referees this would simply lead to the reintroduction of problems
that have been painfully and repeatedly purged over the years.
Our analysis suggests that the bifurcation of mathematics into theoreticaland rigorous communities has partially begun but has been inhibited by theconsequences of improper speculation. Will it continue? Probably it will in any
case, but it should evolve faster and less painfully if the safeguards are adopted.
In the classical areas it will be slower because intelligent speculation must be
based on a mastery of technical detail in previous proofs. In these areas the
12 ARTHUR JAFFE AND FRANK QUINN
framework for speculation is more likely to provide a constructive outlet for
nominally rigorous individuals, whose inspirations exceed their capacity forrigorous proof. Other areas, particularly those involving computer simulations,
are different in that the generation and analysis of data is quite a distinct activityfrom the construction of proofs. In some of these areas specialized theorists
can already be identified and are likely to proliferate.
In any case, the proposed framework provides a context for interactions be-
tween mathematicians and theoretical physicists. Whether or not they become a
permanent fixture in the mathematical community, physicists can be welcomed
as "theoretical mathematicians" rather than rejected as incompetent traditional
mathematicians.
Summary
At times speculations have energized development in mathematics; at other
times they have inhibited it. This is because theory and proof are not just
"different" in a neutral way. In particular, the failure to distinguish carefully
between the two can cause damage both to the community of mathematics and
to the mathematics literature. One might say that it is mathematically unethical
not to maintain the distinctions between casual reasoning and proof. However,
we have described practices and guidelines which, if carefully implemented,
should give a positive context for speculation in mathematics.
Acknowledgment
We wish to thank many colleagues who have made helpful suggestions about
this paper. The first author thanks the John S. Guggenheim Foundation for
a fellowship. Both authors were partially supported by the National Science
Foundation.
References
[A] P. W. Anderson, Concepts in solids, W. A. Benjamin, Inc, New York, 1964.
[D] J. Dieudonné, A history of algebraic and differential topology 1900-1960, Birkhäuser, Basel,1988.
[EH] D. Eisenbud and J. Harris, Progress in the theory of complex algebraic curves, Bull. Amer.
Math. Soc. 21 (1989), 205.
[F] R. P. Feynman, Surely you're joking Mr. Feynman: adventures of a curious character, W. W.
Norton, New York, 1985.
[Fo] J.-M. Fontaine, Valeurs spéciales des fonctions L des motifs, Séminaire Bourbaki, Exposé751, Février 1992, pp. 1-45.
[Gl] J. Gleick, Chaos: making a new science, Viking Penguin Inc., New York, 1987.
[G2] -, Genius: the life and science of Richard Feynman, Pantheon, New York, 1992.
[H] G. Holton, private communication.
[J] A. Jaffe, Mathematics motivated by physics, Proc. Sympos. Pure Math., vol. 50, Amer. Math.
Soc, Providence, RI, 1990, pp. 137-150.
[K] J. Kollar, The structure of algebraic threefolds : an introduction to Mori's program, Bull. Amer.
Math. Soc. 17(1987), 211.
[Kr] S. G. Krantz, Fractal geometry, Math. Intelligencer 11 (1989), 12-16.
[LL] L. Landau and E. Lifshitz, Statistical physics, Oxford Univ. Press, London, 1938.
[M] R. McCormmach, ed., Historical studies in the physical sciences, Princeton Univ. Press,
Princeton, NJ, 1975.
theoretical mathematics 13
[W] A. Weil, Foundations of algebraic geometry, Amer. Math. Soc, Providence, RI, 1946.
[Wi] E. H. Witten, Quantum field theory and the Jones polynomial, Comm. Math. Phys. 121(1989),
351-399.
Harvard University, Cambridge, Massachusetts 02138-2901
E-mail address: [email protected]
Virginia Polytechnical Institute and State University, Blacksburg, Virginia 24061-
0123E-mail address : [email protected]